So dopamine doesn't 'feel' like anything?

Not directly. Dopamine drives motivation and anticipation — the urge to seek, pursue, click, scroll. The actual feeling of pleasure when a reward arrives is mediated by other systems (opioids and endocannabinoids in specific brain regions, called 'hedonic hotspots'). You can have one without the other: drug addiction is largely strong wanting with diminishing liking. Anhedonia is the reverse — you like things you experience but don't want to pursue them.

Is this the basis of how slot machines work?

Yes. Variable-ratio reward schedules (you sometimes win, but never know when) maximise dopamine signaling because every pull is potentially the rewarding one. This is exactly the same mechanism behind social-media feed scrolling — you don't know which post will be interesting, so the next one might be, so you keep scrolling.

What's a 'dopamine detox'?

Not a real biological concept. You cannot deplete or detoxify dopamine through behaviour. What people usually mean — taking a break from highly-stimulating dopamine-driving activities (phones, junk food, porn) — is a real and useful habit reset, but it works through normal habit formation, not through any 'detoxification.' The science term would be 'reducing supranormal reward stimuli.'

Dopamine Isn't the Pleasure Chemical — Here's What It Actually Does

The folk theory

Pop-science version: dopamine is the "pleasure chemical." When something good happens, dopamine is released, you feel pleasure, and you want to repeat the experience. This is why social media, junk food, and addictive drugs hijack your brain.

This story is partly right and very wrong.

What dopamine actually does

Dopamine is the brain's prediction-error signal. When you get an outcome that's better than expected, dopamine neurons fire harder than baseline. When the outcome is as expected, they fire at baseline. When it's worse than expected, they dip below baseline.

The neurons don't care about pleasure per se. They care about whether you're learning that something is worth pursuing. A reward you anticipated produces no dopamine spike — the spike already happened when you anticipated the reward. The actual receipt is just balance-sheet bookkeeping.

This was nailed down by Wolfram Schultz's experiments on monkeys in the 1990s. Dopamine neurons in the midbrain were recorded while monkeys learned that a tone predicted a juice reward. Early in training, the juice produced a dopamine spike. After learning, the tone produced the spike (because it was the unexpected good news now); the juice produced nothing (it was expected). If the juice was withheld unexpectedly, the dopamine signal dipped below baseline at the moment the juice would have arrived — the brain registered "things are worse than predicted."

This is exactly the structure of a temporal-difference learning algorithm in machine learning. The brain runs reinforcement learning on dopamine. The function isn't "feel pleasure"; it's "track when your expectations need updating."

Wanting versus liking

The distinction was sharpened by Kent Berridge's work, beginning in the 1990s. He showed that animals with severely depleted dopamine systems still display the facial expressions of pleasure when given sweet rewards. They like the reward. But they won't work to get it. The pleasure system (liking) is intact; the motivation system (wanting) is shut down.

Conversely, you can artificially over-stimulate dopamine and the animal will pursue rewards compulsively, even ones that have become unpleasant.

So: wanting is dopamine-driven, in the mesolimbic pathway. Liking is mediated by mu-opioid and endocannabinoid systems in small brain regions called "hedonic hotspots." They overlap but are not the same.

In daily life:

A novel beautiful experience = both wanting and liking.
Compulsive scrolling = lots of wanting, little liking.
Severe depression with anhedonia = some liking when reward arrives, no wanting to pursue it.
Drug addiction (long-term) = strong wanting, diminished liking.

The classical "dopamine = pleasure" picture confuses these. Dopamine is closer to craving than to enjoyment.

Why slot machines work

If dopamine fires on positive prediction errors — outcomes better than expected — then uncertainty maximises dopamine. A reliable reward you always get becomes baseline; no spike. A reward you sometimes get and sometimes don't keeps the prediction error large every time you actually get it.

This is the principle behind variable-ratio reinforcement, which is the most addictive reward schedule known. Slot machines, scratch cards, and casino games are built on it. So is most of social media: each post might be the great one, you don't know until you scroll, so you keep scrolling.

Notification systems use the same trick. A buzz might be an important message or junk. The uncertainty itself is the hook. Even when you're on the toilet, ostensibly relaxing, you check your phone — because the variable-ratio reward is always available.

This isn't pathology. It's the standard operation of a normally-tuned reward system being exploited by carefully-engineered apps. Once you see the mechanism, you can sometimes interrupt it.

Habits — the wanting that stays after the liking leaves

A useful framing: a habit is a behaviour where the wanting persists even after the liking has faded.

You first try a new thing — junk food, video game, social platform — and like it. Dopamine spikes from the unexpected reward. You repeat the behaviour, and the dopamine system tags it as worth pursuing. The mental representation of the cue (smell, app icon) becomes associated with the reward.

Repeated enough, the cue itself drives behaviour. You don't need to be enjoying the food, game, or feed; you just need to see the cue. Dopamine spikes on the cue; you act on it almost without deliberation. The original pleasure is no longer required.

This is why long-term smokers describe smoking as "not really enjoyable anymore, I just do it." The liking is gone. The wanting persists. They're stuck in a habit that the dopamine system installed when there was still pleasure attached.

The flip side: building positive habits requires the dopamine system to tag the new behaviour as worth pursuing in the first place. The trick is to make the early reward strong and reliable enough that the dopamine system learns the cue→behaviour→reward loop, after which it persists even without intense conscious motivation.

Practical implications

Some of the things this lens illuminates:

Why anticipation feels different from arrival. The shopping is often more enjoyable than the purchase. Vacation planning is sometimes more thrilling than the trip. Dopamine peaks during anticipation; the actual arrival is anticlimax.
Why "doing the thing" is harder than starting the thing. The dopamine peak from starting (novelty, social signaling) doesn't survive into the grind. Maintaining habits requires the rewards to come during the activity, not just at the start.
Why intermittent reinforcement makes things sticky. A relationship with inconsistent affection, a job where recognition is sporadic, a slot machine — all maximise the dopamine signal precisely because outcomes are uncertain.
Why dopamine "stacking" (caffeine + sugar + a notification) compounds. Each independent reward drives an independent dopamine pulse. You can engineer environments that produce constant low-grade dopamine activity, which feels productive but actually fragments attention.
Why "dopamine fasting" is not what it sounds like. You can't drain dopamine. What you can do is reduce supranormal reward stimuli (phones, junk food, infinite scrolls) and let the system recalibrate. This works through normal habit changes, not detoxification.

What apps and games exploit (and learning could)

Modern engagement design uses dopamine carefully:

Variable reward schedules (uncertain timing of likes, posts, gold drops).
Streaks and counters (a small ongoing dopamine source from maintaining a metric).
Near misses (almost-rewards that feel like they should pay off, generating prediction-error responses without actual payment).
Personalised content (more positive prediction errors per session).

Learning systems can use the same techniques honestly. NerdSip's lesson design — short interactive courses with quizzes, immediate feedback, streaks, and unpredictably interesting facts — uses the same dopaminergic levers that gambling apps do, applied to acquiring knowledge instead of losing money. The mechanism is neutral. What it gets attached to is the design choice.

The takeaway

Dopamine isn't pleasure. It's a prediction-error signal that builds motivation toward whatever produces unexpected good outcomes. Liking — actual enjoyment — runs on a separate system. You can have wanting without liking (addiction, compulsive scrolling), liking without wanting (anhedonia), or both at once (a novel pleasure). Slot machines and social media exploit the variable-ratio schedule that maximises dopamine signaling. Habits are the residue of dopaminergic learning that persists after the original pleasure has faded. The whole field of behavioural economics, design, and addiction medicine sits on this distinction.